A construction of processes with one dimensional martingale marginals, based upon path-space Ornstein-Uhlenbeck processes and the Brownian sheet (Q2654728): Difference between revisions

Let \(E=C([0,1],{\mathbb R})\) with the Wiener measure and the usual Brownian \(B\) and filtration \({\mathcal B}\), \(E^{2}=C([0,1]^{2},{\mathbb R})\) with the Brownian sheet \(W\) and filtration \({\mathcal W}\). Let \(B^{\diamond}\) be another Brownian, independent of \(B\) and \(W\). For every \(\Phi \in L^{2}(E)\) the authors consider the processes \(\Phi_{t}^{\sharp }= E_{B^{\diamond}}(\Phi (B_{t\cdot }+(1-t)^{1/2}B^{\diamond}))\) (\({\mathcal B}_{t}\) adapted and \(L^{2}\) continuous) and \(\Phi_{t}^{m}=E_{B^{\diamond}}(\Phi (W_{\cdot ,t}+(1-t)^{1/2}B^{\diamond}))\) (a \({\mathcal W}_{1,t}\) martingale). Other expressions of these processes in terms of the chaos representation of \(\Phi\) are proved. The authors show as result of main interest the fact that, for every \(t\), \(\Phi_{t}^{\sharp }\) and \(\Phi_{t}^{m}\) have the same distribution. Then they define the following classes of \(\Phi\)'s: \({\mathcal I}\) (the processes \(\Phi^{\sharp }\) and \(\Phi^{m}\) have the same distribution), \({\mathcal M}\) (\(\Phi^{\sharp }\) is a \({\mathcal B}\)-martingale), \({\mathcal V}\) (\(\Phi^{\sharp }\) is continuous with finite variation), \({\mathcal S}\) (\(\Phi^{\sharp }\) is a semimartingale). Examples of \(\Phi\)'s in \({\mathcal I}\cap {\mathcal M}\): \(f(B_{t})\) for \(f\) in \(L^{2}\) of null mean, gaussian with variance in \((0,1)\) and \(\varphi (r,B_{r})\) for \(r \in [0,1]\), \(\varphi '_{t}=-\varphi ''_{_{x}2}/2\). For \(R_{t}\Phi = \Phi_{t}^{\sharp }\), \(R_{_{e}-t}\) is a semigroup on \(L^{2}(E)\), the domain of its infinitesimal generator \({\mathcal A}\) is included in \({\mathcal V}\) and, for \(\Phi \in \text{Dom}{\mathcal A}\), \(\Phi^{\sharp }\) has square integrable variaton, hence is not in \({\mathcal M}\) if not constant in \(t\). Also the local time in \(a\) at \(r\) is in \({\mathcal V}\). A core of \({\mathcal A}\) is defined in terms of the chaos representation. \(\Phi = \int_{_{0}}^{^{1}}F_{t}^{\sharp }h(t)dt\) is in \(\text{Dom}{\mathcal A}\) for \(EF=0\) and absolutely continuous \(h\) with \(\int_{_{0}}^{^{1}}u^{3/2}|h`(u)|<\infty\), even such \(\Phi\) with \(h \in C^{1}\) are dense in \(L^{2}\). \({\mathcal M}\) is the closure in \(L^{2}(E)\) of the set \({\mathcal N}\) of all \(\Phi =a+\int_{_{0}}^{^{1}}F_{u}^{\sharp }dB_{u}\) with \(F\in L^{2}(E)\), but \({\mathcal M}\neq {\mathcal N}\) as is seen from the example \(f(B_{t})\) in \({\mathcal M}\), which is in \({\mathcal N}\) if and only if \(f\) is absolutely continuous on \({\mathbb R}\) with a square integrable derivative. Even, for \(r\in [0,1]\), \(\Phi = \int_{_{0}}^{^{r}}F_{u}^{\sharp }dB_{u}\), \(\Phi_{t}^{\sharp }\) is a \({\mathcal B}_{tr}\) martingale and such \(a+\Phi\) have as closed linear hull \(L^{2}(E)\). That \({\mathcal I}\subset {\mathcal M}\) is false is seen studying the \(\Phi = \int_{_{0}}^{^{1}}h(u)dB(u)\), while that even \({\mathcal N}\subset {\mathcal I}\) is false is seen with the aid of \(\Phi = \int_{_{0}}^{^{1}}B_{au}dB_{u}\), \(a\in (0,1)\). The paper finishes with a sufficient condition for \(\Phi \in {\mathcal S}\) in terms of the chaos representation.